Shade-GAN-Based 3D Reconstruction of Missing Children Faces
Qingyao Li, Ying Kin Yu, Ming Gao Yan · 2023
Children lost from their families is a heartbreaking problem, and the process of finding a lost child is frequently challenging. The more significant issue is that families of lost children frequently have very limited image resources at their disposal. Therefore, we strive to find a way to generate more available resources with limited resources, i.e., unsupervised 3D reconstruction of human faces. In this paper, we use the Shade-GAN (Generative Adversarial Network, GAN) model for the 3D reconstruction of children's faces to achieve more realistic 3D perceptual image synthesis through the interaction of illumination and shape, simulated illumination and shadows with various illumination conditions to achieve multiple illumination constraints, and an efficient volume rendering strategy. We applied two sets of parametric models of shade-GAN on the UTK-20 dataset containing 4872 images and analyzed the degree of accuracy of the obtained results, and in the end, we concluded the applicability of shade-GAN models in the field of finding lost children and ways to improve the performance of what we obtained.